Agronomy & Horticulture -- Faculty Publications Agronomy and Horticulture Department 3-18-2019
Development of a nitrogen recommendation tool for corn
Development of a nitrogen recommendation tool for corn
considering static and dynamic variables
considering static and dynamic variables
Laila A. Puntel
University of Nebraska-Lincoln, [email protected] Agustin Pagani
Clarion Inc., Buenos Aires Sotirios V. Archontoulis
Iowa State University, [email protected]
Follow this and additional works at: https://digitalcommons.unl.edu/agronomyfacpub
Part of the Agricultural Science Commons, Agriculture Commons, Agronomy and Crop Sciences Commons, Botany Commons, Horticulture Commons, Other Plant Sciences Commons, and the Plant Biology Commons
Puntel, Laila A.; Pagani, Agustin; and Archontoulis, Sotirios V., "Development of a nitrogen
recommendation tool for corn considering static and dynamic variables" (2019). Agronomy & Horticulture -- Faculty Publications. 1249.
https://digitalcommons.unl.edu/agronomyfacpub/1249
This Article is brought to you for free and open access by the Agronomy and Horticulture Department at
DigitalCommons@University of Nebraska - Lincoln. It has been accepted for inclusion in Agronomy & Horticulture -- Faculty Publications by an authorized administrator of DigitalCommons@University of Nebraska - Lincoln.
Development of a nitrogen
recommendation tool for corn
considering static and dynamic variables
Laila A. Puntel,
1Agustin Pagani,
2and
Sotirios V. Archontoulis
11 Iowa State University, 3401 Agronomy Hall, Ames, IA 50014, USA 2 Clarion Inc. Nueve de Julio, Buenos Aires, Argentina
Current address for L.A. Puntel — Department of Agronomy and Horticulture, 102D Keim Hall, University of Nebraska–Lincoln, Lincoln, NE 68583-0915, USA;
email [email protected] Abstract
Many soil and weather variables can affect the economical optimum nitrogen (N) rate (EONR) for maize. We classified 54 potential factors as dynamic (change rap
-idly over time, e.g. soil water) and static (change slowly over time, e.g. soil organic matter) and explored their relative importance on EONR and yield prediction by analyzing a dataset with 51 N trials from Central-West region of Argentina. Across trials, the average EONR was 113 ± 83 kg N ha−1 and the average optimum yield was 12.3 ± 2.2 Mg ha−1, which is roughly 50% higher than the current N rates used and yields obtained by maize producers in that region. Dynamic factors alone ex
-plained 50% of the variability in the EONR whereas static factors ex-plained only 20%. Best EONR predictions resulted by combining one static variable (soil depth) to
-gether with four dynamic variables (number of days with precipitation>20 mm, res
-idue amount, soil nitrate at planting, and heat stress around silking). The resulting EONR model had a mean absolute error of 39 kg N ha−1 and an adjusted R2 of 0.61.
Interestingly, the yield of the previous crop was not an important factor explaining EONR variability. Regression models for yield at optimum and at zero N fertilization
1
Published in European Journal of Agronomy 105 (2019), pp 189–199 doi 10.1016/j.eja.2019.01.003
Copyright © 2019 Elsevier B.V. Used by permission.
Submitted 16 June 2018; revised 10 January 2019; accepted 11 January 2019; published 18 March 2019.
rate as well as regression models to be used as forecasting tools at maize planting time were developed and discussed. The proposed regression models are driven by few easy to measure variables filling the gap between simple (minimum to no inputs) and complex EONR prediction tools such as simulation models. In view of
increasing data availability, our proposed models can be further improved and de-ployed across environments.
Keywords: Optimum nitrogen rate recommendation, Corn, Decision support tools, Precision agriculture, Argentina, Site-specific management
1. Introduction
Maize production in Argentina has doubled over the last five years from 20 to 40 million tones (AMIS, 2018). About 80% of the grain produced is ex
-ported annually, making this region important in the global maize mar
-ket (Alexandratos and Bruinsma, 2012; Andrade and Satorre, 2015). The in -creased corn production has been mainly associated with an increase in
farmable area rather than productivity. The Central-West region’s average
corn yield is 7.6 Mg ha−1, which is considerably below the potential yield of
16 Mg ha−1 (Andrade and Satorre, 2015). Poor crop management, in partic -ular nitrogen (N) fertilization, is the major reason for the yield gap. The av -erage N application rate to corn in this region is about 61 kg N ha−1 (Bolsa de Cereales, 2018), which is about half of the US Corn Belt average N rate (Sawyer et al., 2006). Increasing N fertilization results in increased produc -tion costs that depending in the year, soil type, and landscape posi-tion this
investment may or may not pay off due to the multiple factors affecting the economic optimum N rate (EONR, Puntel et al., 2018).
A better understanding and predictability of the EONR variability in this region could result in higher yields and profits while reducing environmen
-tal impacts (Aparicio et al., 2008). However, there is little research on EONR in Central-West Buenos Aires compared to other regions such as the US Corn Belt (Scharf and Lory, 2006; Dhital and Raun, 2016). Interestingly, the vast majority of published studies on EONR have focused on the effects of
individual factors such as soil properties or precipitation as opposed to the
compound effect of multiple factors on EONR (Mamo et al., 2003; Basso et al., 2001, 2013; Albarenque et al., 2016). To our knowledge, in Argentina
the most common N recommendation methods is the soil N test at 60 cm
depth before planting (Ruiz et al., 2001) or at 6th leaf stage at 30 cm (Sainz Rozas et al., 2000). However, other N recommendation methods have been
evaluated in Argentina such as the N balance approach, stalk and plant N
test (Sainz Rozas et al., 2001), crop sensors and remote sensing technolo
2001). The majority of these tools does not account for spatio-temporal dy
-namics such as soil available water content at sowing (Gregoret et al., 2011; Coyos et al., 2018).
In general, factors affecting EONR and yield can be broadly classified into dynamic variables (ones that change quickly within a growing season such as precipitation) and static variables (ones that change slowly and do not vary within a growing season such as soil organic matter, SOM). Today’s N rec
-ommendation tools are based on: 1) dynamic factors such as soil nitrate-N (Rozas et al., 2000; Shapiro et al., 2008), 2) static factors such as soil texture (Tremblay et al., 2011), 3) both static and dynamic factors via crop and soil modeling (Banger et al., 2017), 4) yield expectations (Stanford et al., 1973) and 5) multiyear- site average yield response to N curves (e.g. MRTN, Saw
-yer et al., 2006) without considering other factors.
Regarding the static variables, some studies have reported relationships between yield or EONR with soil texture and/or SOM (Gregoret et al., 2006; Peralta et al., 2013; Puntel et al., 2017). Soil texture and SOM affect water holding capacity and N cycling, thus soil N supply and crop N uptake (Sog
-bedji et al., 2001). However, measuring these static variables at a fine spatial resolution is expensive and labor intensive. To address this issue more acces -sible variables such as topography and apparent soil electrical conductivity
(ECa) are used (Kitchen et al., 2003; Shaner et al., 2008). The ECa is correlated with soil water content (Brevik et al., 2006), soil compaction (Kravchenko and Bullock, 2000), and salinity (Heiniger et al., 2003) and it is a very good indi
-cator of soil texture (King et al., 2005). Topography, derived terrain parame -ters, and ECa have been used with varying degrees of success to determine
areas with contrasting yields (Chang et al., 2004) and differential yield re
-sponse to N (Jaynes, 2011).
Regarding the dynamic variables, several studies have indicated the im -portance of soil moisture, soil nitrate, rainfall and temperature on crop yields
and EONR in rainfed regions (Andrade et al., 1993). For example, Hergert et al. (1995) and Kitchen et al. (2005) illustrated the importance of spatial and temporal dynamics of soil nitrate on EONR. Ordóñez et al. (2015) and Edreira and Otegui (2013) quantified heat stress effects on corn yield and N uptake. Soil temperature and moisture affect soil N mineralization and residue de
-composition and thus crop N availability and EONR (Andraski et al., 2000; Cabrera et al., 2005). Shallow groundwater dynamics have been found to positively affect yield in dry seasons and negatively affect yield in wet sea
-sons (Jaynes, 2012; Nosetto et al., 2009) while also affecting environmental N losses (Dinnes et al., 2002) and thus the EONR.
Understanding which dynamic and static factors or synergic relationships
contribute the most to spatial (between landscape positions) and tempo
2015). Thus far, no study has been conducted to compare the relative im
-portance of different static and dynamics factors on EONR and corn yield. A targeted experimental approach is needed in which several variables are
simultaneously measured to identify the most important ones for further
emphasis. To shed light on this important knowledge gap, we analyzed a dataset with 51 N response trials from Central-West Buenos Aires, Argen
-tina in which numerous static and dynamics variables were measured. Our specific objectives were to: 1) identify the range of yield and EONR variabil
-ity in this region, 2) quantify the relative importance of dynamic and static factors on yield and EONR, and 3) to synthesize gained knowledge and de
-velop a predictive N model to aid site-specific N management. 2. Material and methods
2.1. Experimental sites and design
Fifty-three N rate trials were conducted at contrasting landscape positions and soil types, on five fields located in Nueve de Julio, Buenos Aires, Argen
-tina during five growing seasons: 2012–2013 (season 1; 10 trials), 2013–2014 (season 2; 10 trials), 2014–2015 (season 3; 13 trials), 2015–2016 (season 4; 11 trials), and 2016–2017 (season 5; 9 trials). Soils were coarse-loamy, ther
-mic Typic Hapludolls representing the most productive areas of the fields,
coarse-loamy, thermic Entic Hapludolls mostly representing sandy hills with
low productivity, and Thapto-argic Hapludolls corresponding to shallow soils due to the presence of a clay pan layer at varying depths. These soils are rep
-resentative of the Central-West Buenos Aires Province and other provinces in Argentina. In addition, the area is influenced by a fairly shallow water ta
-ble that responds rapidly to rain (Mejia et al., 2000). Due to severe flood -ing two trials were not harvested reduc-ing the number of N trials used in the analysis to 51.
The N-trials were established in representative landscape positions with
the goal to generate variability in elevation, ECa, soil nitrogen, soil
proper-ties, and previous crop productivity (e.g. summer soybean-corn or double crop spring wheat/summer soybean-corn rotations). As an example, Fig
-ure S1 illustrates the position of six N trials. Each N-trial was a small-plot randomized complete block design with three replications. Seven N rates (0, 25, 50, 100, 150, 200, and 250 kg ha−1) were applied as broadcasted urea between planting and third corn leaf stage (V3 stage; Abendroth et al., 2011). Plot size was 9m long and 2.8m wide. According to soil analy
-sis, only phosphorus and sulfur were below optimum values, thus, fertiliz -ers were applied according to soil analysis and local recommendations to
ensure nutrient sufficiency. All trials had a final plant population between
65,000 and 80,000 plants ha−1 and row spacing was either 0.75m or 0.52
m. Commercial corn hybrids with a relative maturity of 120–125 days were
planted around September 25th±7 days. Fields were managed without till -age. Pest, diseases, and weeds were adequately controlled to ensure op-timal growing conditions.
2.2. Measurements and data processing
Grain yields were determined by collecting ears from the center two rows of each plot (5m length). Grain moisture was measured, and final yields were
adjusted to 14% moisture content. Additional measurements were taken
from each N trial to explain yield response to N including: SOM, texture,
gravimetric water content, ECa, elevation, soil and water table depth, amount
of residue from the previous crop, and hourly weather data. These measure
-ments and subsequent calculations were classified into static and dynamic explanatory variables, which are described below and summarized in Table 1.
2.2.1. Static variables
Static variables are relatively constant over time and measured once per trial
(Table 1). Soil apparent electrical conductivity at 30 and 90 cm depth was measured on transects, approximately 20m apart using a Veris model 3100 sensor cart system (Veris Technologies, Salina, Kansas, USA). The ECa sur -veys were conducted before planting or after harvest. Elevation data was
obtained by a dual frequency RTK system (Trimble 5700, USA) connected to
the EC Veris surveyor. Both ECa and elevation data points were interpolated
using ArcGIS (ESRI, 2018, Redlands, CA, USA) and R software (R Core Team, 2018) using ordinary kriging in a regular 3-m grid (Figure S1).
Landscape characteristics were determined by primary and secondary
terrain attributes (Moore and Grayson, 1991; Wilson and Gallant, 2000). We used primary attributes such as elevation, relative elevation (Rel_elev), slope, and plan curvature (pcurv) that were derived directly from digital elevation models (DEM) (Figure S1, Table 1). Secondary attributes such as specific catchment area (SCA) were derived from a combination of primary attri
-butes. Digital terrain analysis was performed using the GRASS 7.0.5 (Geo
-graphic Resources Analysis Support System, grass.osgeo.org) and ArcGIS
10.5 software packages.
The r.param.scale function in GRASS was used to calculate plan curva
-ture. The digital elevation model was processed using a neighborhood of 3 by 3 cells (equivalent to an analysis scale of 13.9 m). The slope gradient and a moving-window version of relative elevation (REL; Miller, 2014) were cal
Table 1. Description of static and dynamic explanatory variables.
Acronym Explanation Unit Observed range
Static variables (change slowly over time or do not change)
OM_20 Soil organic matter (0-20 cm depth) % 2-4
P_Bray_20 Available Bray phosphorus (0-20 cm depth) ppm 6-51
EC_20 Electrical conductivity (0-20 cm depth) ds/cm 47-124
pH_20 pH (0-20 cm depth) – 5-6
Sand_20 Sand content (0-20 cm depth) % 36-84
Silt_20 Silt content (0-20 cm depth) % 12-53
Clay_20 Clay content (0-20 cm depth) % 3-16
ECa_30 Soil apparent electrical conductivity (0-30 cm depth) ds/m 3-19 ECa_90 Soil apparent electrical conductivity (0-90 cm depth) ds/m 4-28
Elev Elevation as meters above the sea level m 71-89
Rel_elv Relative elevation using the middle elevation as the reference % −0.22-0.29 (REL, Miller et al., 2014)
planc Plan curvature deg −0.28-40
Slope Slope of the field %_raise 0-5
SCA Specific catchment area pixel value 0-486
MO (20-60) Soil organic matter content (20-60 cm depth) % 1-3
Sand (20-60) Sand content (20-60 cm depth) % 36-84
Silt (20-60) Silt content (20-60 cm depth) % 12-48
Clay (20-60) Clay content (20-60 cm depth) % 5-20
Soil_depth Soil depth m 0.6-2
Dynamic variables (change fast over time)
Residue_amount Amount of residue from previous crop at planting kg/ha 3636-10909
Residue_CN Quality of residue as Carbon to N ratio – 30-75
Previous Yield Yield of the previous crop kg/ha 2000-6000
N_20 Nitrate content (0-20 cm depth) kg/ha 19329
N (0-60) Nitrate content (0-60 cm depth) kg/ha 28-116
Water table Water table depth cm 30-450
SW_20 Soil water content (0-20 cm depth) mm 21-72
SW (20-60) Soil water content (20-60 cm depth) mm 30-177
FC_20 Soil water as a % of field capacity at 20 cm % 43-237
FC (20-60) Soil water as a % of field capacity at 20 to 60 cm % 71-265
FC_w Soil water as a % of field capacity 0 to 60 cm % 65-253
Max_20 Soil water as a % of saturation point at 20 cm % 23-82
Max (20-60) Soil water as a % of saturation point at 20 to 60 cm % 35-102
Max_w Soil water as a % of saturation point at 0 to 60 cm % 34-93
SW_sum Soil water content (0-60 cm depth) mm 57-277
Events_P-S Number of rain events from planting to silking days 9-28
Amount_P-S Cumulative rain from planting to silking mm 202-528
Events_P-S_20 Number of days with rain > 20mm from planting to silking days 2-13 Amount_P-S_20 Cumulative rain (only > 20 mm) from planting to silking mm 98-428 Events_S Number of rain events around silking (±1.5 weeks) days 1-8
Amount_S Cumulative rain around silking (±1.5 weeks) mm 25-176
Events_S_20 Number of days with rain > 20mm around silking days 0-7 Amount_S_20 Cumulative rain (only > 20 mm) around silking mm 0-150 Events_H-P_E Number of rain events from harvest to planting days 6-25 Events_H-P_20 Number of rain events (> 20 mm) from harvest to planting days 2-12
Amount_H-P_E Cumulative rain from harvest to planting mm 140-866
Amount_H-P_20 Cumulative rain events (> 20 mm) from harvest to planting mm 98-810 Events_P-H_20 Number of rain events from planting to harvest days 11-60 Events_P-H_20 Number of rain events (> 20 mm) from planting to harvest days 2-18
Amount_P-H_20 Cumulative rain from planting to harvest mm 296-840
Amount_P-H_20 Cumulative rain (only > 20 mm) from planting to harvest mm 78-672 Temp_P-H_35 Number of heat days (daily temp > 35 °C) around silking days 2-13
Temp_S_35 Number of heat days from planting to harvest days 1-5
Radiation_S Classification of radiation around silking *** – 1-2
models (available at http://www.geographer-miller.com/relief-analysis-tool-box ). Plan curvature is perpendicular to the direction of the maximum slope. A positive value indicates that the surface is sidewardly convex at that cell.
A negative value indicates the surface is sidewardly concave at that cell. A
value of zero indicates the surface is linear (Figure S1). Profile curvature re
-lates to the convergence and divergence of water flow across a surface. The SCA was calculated using Spatial Analyst Hydrology Tools in ArcGIS 10.5. The SCA is defined as the area of land upslope of a width of contour, divided by the contour width. This is a commonly used quantity in hydrol
-ogy to describe complex terrain for analyzing water flow on hill slopes; it can be a surrogate for water discharge per unit flow width.
Soil organic matter was determined by the combustion method (Wang and Anderson, 1998) and texture by the pipette method (Soil Survey Staff, 2014). Ten cores were taken per block from 0 to 20, 20–60, 60–100 cm depth in most of the experimental sites to determine SOM and texture. For the first two seasons, samples were collected from 0 to 20, 20–40, and 40–60 cm depth (Sites 1–10). Data was combined into top (0–20 cm) and subsoil data (20–60 cm). Data below 60 cm was not used in the analysis. Effective soil depth, which stands for the depth of the pretocalcic horizon and/or clay -pan layer, was measured manually with a 1.2-meter soil probe in each plot.
Soil depth was as shallow as 60 cm for some of the experimental sites. Us
-ing SOM, texture data, and Saxton and Rawls (2006) pedotransfer functions, we calculated field capacity (FC) and saturation point (SAT) for the top and
subsoil layers.
2.2.2. Dynamic variables
Variables that change rapidly over time were classified as dynamic (Table 1). Gravimetric soil water and nitrate-N contents at 0–20 cm and 20–60 cm were measured at planting. Soil water content was expressed as a percent of FC and SAT. Nitrate-N content was determined using the phenoldisulphonic acid method (Davis, 1916) and expressed as kg of N per ha by layer. The to
-tal nitrate-N amount across the 0–60 cm profile was used as an explanatory variable in this analysis. Water table depth at each trial was manually mea
-sured (from installed wells) around planting. Hourly precipitation, radiation, and temperature data was obtained from nearest weather station (< 5 km) to the experimental sites. Precipitation was accumulated for the following periods: from harvest of the previous crop to planting of corn (amount_H-P), from planting to silking (amount_P-S),±1.5 weeks around silking (criti
-cal period; amount_S), and from planting to harvest (amount_P-H). For the same time periods, we calculated and used as explanatory variables the
number of rain events greater than 0mm and greater than 20mm per day
(events_H-P, events_S, H, events_H-P_20, events_S_20, events_P-H_20). The value of 20mm day−1 was arbitrarily chosen in this analysis to
account for possible N leaching. The number of days with air temperature below 10 °C during the growing season, the number of days with air tem
-peratures above 35 °C around maize’s critical period (±1.5 weeks around silking), and over the entire growing season were counted (Temp_P-H_10, Temp_P-H_35, Temp_S_35, respectively). Lastly, we calculated the radiation sums around the critical period (Radiation_S) as a proxy of crop growth rate that is known to affect kernel number.
Grain yield of the previous crop was estimated from yield monitor data at each experimental site. When yield maps were not available, yields of
the previous crop at each site were estimated based on image analysis and
yield records from the farmer. The amount of residue and its carbon (C)-to-nitrogen ratio (C:N) was directly measured in season 4 (2015-16) in each
block from an area of 1m2. A subsample of the residue was analyzed for C, N, and C:N ratio using the dry combustion method (LECO, 2008). Residue
amount and quality for the other seasons were estimated from previously
published grain yield and residue C:N relationships for Argentina (Melchi
-ori et al., 2014). 2.3. Data analysis
The relationship between yield and N rate was described from the quadratic and quadratic-plus-plateau models using R software (R Core Team, 2018). Models were deemed significant at p < 0.05 and the equations with the
smallest sums of squares and largest R2 were selected (Table S1). The EONR and yield at the EONR (YEONR) was calculated from the N response equa
-tions by setting the first derivative of the fitted response curve equal to a historical price ratio of 10:1 N: corn grain price (US$ kg−1 N: US$ kg−1 grain) ratio (Cerrato and Blackmer, 1990; Bullock and Bullock, 1994; Pagani et al., 2008). This price ratio is about half of what is used in the USA (5.6:1; Puntel et al., 2016). A sensitivity analysis of the effect of price ratios (5.6 and 14 vs 10 N: corn grain price) on EONR for three experiments is shown in Figure S6. Although optimum N rates would be slightly different if different prices were used, the optimal N rate is relatively insensitive to shifts in prices (Ba
-ethgen et al., 1989; Pagani et al., 2008). Maximum yield response to N was calculated as the difference between YEONR and Yield_N0.
Relationships between explanatory variables (Table 1) were explored us
-ing Pearson correlation, principal component analysis, and cluster-ing (Fig
-ure S2). We excluded highly auto-correlated variables from subsequent re -gression analysis. Plant population and row spacing was considered random
in the statistical analysis. The remaining static and dynamic variables were used to develop regression models for EONR, YEONR, and Yield_N0. The efficient branch-and-bound algorithm method within the leaps R package
a linear model for each sub-model individually to obtain the selection
cri-teria. The best model was selected based on adjusted R2 and k-fold (leave-one-out) cross validation error. Finally, we calculated performance indexes such as mean absolute difference (MAE) and root mean squares (RMSE, see equations in Archontoulis and Miguez, 2015).
We developed two types of regression models (Table 2). The first one,
hereafter referred to as the full model, made use of all available
informa-tion from harvesting of the previous crop to harvesting of the next crop. The
second, hereafter referred to as the reduced model, made use of
informa-tion from harvesting of the previous crop to planting time of the next crop. To determine the relative importance of static and dynamic variables within the regression models we used the simple unweighted averages (lmg) method (Gromping, 2006). This metric decomposes R2 into absolute contri-butions of different factors that sum to the total R2. The advantage of this method over simpler metrics is that the lmg is based on sequential R2, while
accounting for the dependence on orderings using simple unweighted
aver-ages. This analysis was performed for the full and reduced model separately. 2.3.1. Preliminary model testing
We used an independent set with three N trials from the 2017-18 (Season 6) growing season to test the predictive capacity of both full and reduced
models. This season had a severely dry summer (244mm of rain from plant
-ing to harvest; with only 30mm around silk-ing). The soils had a SOM from 2.2 to 3.5% and a sand content from 52 to 70%. The previous season’s crop
was a double crop of spring wheat/summer soybean. All model inputs were calculated as described in previous sections.
Table 2. Mean absolute error for economic optimum nitrogen rate (EONR; units:
kg N ha−1), yield at EONR (YEONR; Mg ha−1), and yield at nitrogen zero (Yield_N0;
Mg ha−1) for regression model predictions using information from previous harvest until planting (reduced model) and using information available from previous crop harvest to next harvest (full model).
kg ha-1
Season In-season prediction (reduced model) At planting prediction (full model) EONR YEONR Yield_N0 EONR YEONR Yield_N0
1 49 1.3 0.7 52 1.2 1.2 2 45 1.6 1.3 56 1.7 1.3 3 40 1.4 1.1 37 1.7 1.6 4 32 1.1 0.9 42 1.4 1.3 5 24 2 1.3 35 1.8 1.4 All seasons 43 1.2 1.3 39 1.1 1.0
3. Results
3.1. Spatial and temporal variability of the explanatory variables Season 1 was the wettest (23% above regional average precipitation) and Season 2 was the driest (7% below average precipitation; Figure S3). During the critical period of silking, the number of precipitation events (> 0 mm)
was the highest in Season 5 and the lowest in Season 2.
Soils had on average 80 kg N ha−1 at planting time in the dry season (Sea -son 2), this was 60% more than the average at the wet Sea-sons, 1, 4 and 5 (Fig. 1). Soil nitrate-N varied the most during the extremely wet and dry sea
-sons (Fig. 1). Season 2 also had the highest variation in soil water content expressed as a percent of field capacity between landscape positions (Fig. 1). The depth to the water table varied from 30 to 300 cm across seasons and treatments (Fig. 1). Season 5 had the highest number of days with tem
-peratures below 10 °C and Season 2 had the greatest number of days with temperatures above 35 °C (45 and 13 days, respectively; data not shown).
The range of values observed for the static values are reported in Table 1. Some static variables were correlated with one another. We found that the estimated terrain parameters correlated well with soil texture, SOM, and ECa_90 (Figure S4). Soils with high slopes were characterized by low SOM and high sand content whereas high SCA (specific catchment area) was as
-sociated with areas of low sand content. As ECa_90 increased, SOM, and SCA increased (Figure S4).
3.2. Temporal and spatial variability of EONR and yields
Across the 51 N-trials, the EONR varied from 0 to 260 kg N ha−1 with a mean of 113 ± 81 kg N ha−1 (Fig. 2). The EONR values were above the 51 trial mean in 90% of the cases in Season 1, and only 13% of the cases in Season 5 (Fig. 2). The Yield_N0 was below the 51 trial mean value (9.5 Mg ha−1) in all cases in Season 1 and above the mean value in 87% of the cases in Season 5 (Fig. 2). The mean YEONR was 12.2 Mg ha−1 and its distribution across the 51
tri-als was skewed to the right. In 3 of the 5 seasons, the spatial variation of
EONR and YEONR was higher than their temporal variation (CV of 72% ver
-sus 27%, respectively, Fig. 2). The variability in Yield_N0 was higher across years (27%) than within fields (21%). The YEONR varied less compared to EONR and Yield_N0 (Fig. 2).
3.3. Correlations between EONR, yield and explanatory variables During the wettest season (Season 1), highly productive areas (fine texture) had an average EONR of 213 kg N ha−1, whereas low productive areas with
the sandiest soils (sand content>65%) also had high EONR (˜ 195 kg N ha−1, Fig. 1 and 2). Field areas with high sand content (> 65%) and low SOM (< 2.5%) resulted in high temporal variability in yield and EONR (Fig. 2). We found no relationship between EONR and YEONR (Fig. 3). However, the dif
-ference between YEONR and Yield_N0 that is the yield response to N was highly correlated with the optimal N rate (RAdj2 = 0.91; Fig. 3). In very wet sea -sons, the yield response to N was double that of the dry seasons. The dif
-ferences in yield response to N were 3- fold higher in fine than coarse tex
-tured soils in dry and normal seasons. In general, the EONR, YEONR, and Yield_N0 tended to increase as SOM increased and sand content decreased (Figure S5). The EONR and Yield_N0 was significantly correlated with pre
-cipitation from planting to silking (R2 from 0.20 to 0.57), but the magnitude of the response was associated with soil texture and SOM (data not shown). Fig. 1. N-nitrate content from 0 to 60 cm depth, water as percent of field capacity from 20 to 60 cm soil depth (FC), soil water from 0 to 60 cm depth (SW_sum), soil organic matter at 20 cm depth (OM_20), water table depth at planting, and sand content at 20 cm depth (Sand_20).
Fig. 2. Observed economic optimum nitrogen rate (EONR), yield at EONR (YEONR), and yield at nitrogen zero (Yield_N0, right panels) and frequency distribution (right panels). Horizontal dashed lines represent the overall mean value. The coefficient of variation (CV%) for each season is provided.
Fig. 3. Economic optimum N rate (EONR) as a function of yield at EONR (YEONR), yield at N zero (Yield_N0), and the difference between YEONR and Yield_N0.
3.4. Relative importance of factors and model development
The variability in EONR and Yield_N0 was best explained by dynamic factors while the variability in YEONR was best described by static variables (see RAdj2 values in Fig. 4). Prediction accuracy substantially increased when we com
-bined dynamic and static variables (RAdj2>0.60, Fig. 4).
From a total of 54 static and dynamic variables examined in this study (Table 1), four dynamic variables (precipitation, heat stress, nitrate- N at planting, residue amount) and one static variable (soil depth) were consid
-ered in the EONR full model based on their importance (Table 3). The re
-sulting model explained 61% of the variability in EONR with a MAE of 39 kg
N ha−1 (Fig. 2). The number of precipitation events (> 20 mm) from plant
-ing to silk-ing was found to be the most important variable and heat stress
Fig. 4. Diagram of regression models for the economic optimum nitrogen rate (a), yield at the EONR (b), and Yield at N0 (c) using static variables, dynamic variables,
and a combination of both. Predicted versus observed values for the full models
are shown. Diagonal dashed lines are 1:1. The relative importance of static and dy
-namic variables included in the final model as shown. All acronyms are explained in Table 1. Adjusted coefficient of determination (R2), mean absolute error (MAE), and
the least important variable in the full EONR model (see variance analysis
in Fig. 4). For YEONR and Yield_N0 the best single were the residue amount and the number of rain events (> 20 mm) from planting to harvest, respec
-tively (Fig. 4; Table 3).
We also re-ran the same analysis with a reduced amount of data (from previous crop harvest to planting of the new crop) and developed three ad -ditional models that can be used for forecasting purposes at planting time.
The models and parameter values are listed in Table 3. In the reduced
mod-els, information on soil water and soil nitrate-N at planting time became
highly important in the EONR and Yield_N0 prediction (Fig. 5 and Table 3). Residue amount and Rel_elev were the most important variables for the re-duced YEONR model (Fig. 5).
The prediction accuracy of the reduced models was similar for the EONR,
YEONR, and Yield_N0 (RAdj2>0.55, Fig. 5). On average, the full EONR model Table 3. Full and reduced regression models for the economic optimum nitrogen rate (EONR;
units: kg N ha−1), yield at EONR (YEONR; Mg ha−1), and yield at nitrogen zero (Yield_N0; Mg
ha−1). Acronyms explanation and units are provided in Table 2. An Excel version of these mod
-els is provided as supplementary materials.
Type Model
Full EONR = 355 – 163*Soil depth + 22.4*Events_P-S_20 (uses data from harvesting of the – 10.3*Events_P-H_20 – 1.9*N(0-60)
previous crop to harvesting of + 17.1*Temp_S_35 + 0.0196*Residue_amount the next crop)
YEONR = 18.1 – 11.6*Rel_elev – 0.18*Events_P-H_20 – 0.008*SCA + 0.013*Amount_S – 3.8*Soil_depth + 0.0058*Residue_amount
Yield_N0 = 4.96 – 0.253*Events_P-H_20 + 0.05*N(0-60) – 12.56*Rel_elev – 0.008*SCA
– 0.56*Temp_S_35 + 0.2*Temp_P-H_10
Reduced EONR = 390 – 145*planc – 1.96*N(0-60) – 0.83*FC(20-60) (uses data from harvesting of the previous + 11.1*Events_H-P_20 – 120*Soil_depth
crop to planting time of the next crop) + 0.017*Residue_amount
YEONR = 13.8 – 11*Rel_elev – 0.01*SCA + 0.003*Amount_H-P + 0.016*SW_sum + 0.0005*Residue_amount – 3.7*Soil_depth Yield_N0 = 1.1 –13.7*Rel_elev – 0.010*SCA
+ 0.05*N(0-60) + 0.006*Water table + 0.010*FC(20-60) + 0.024*SW_sum
outperformed the reduced models’ prediction accuracy by 16% (Table 2).
The full model predicted EONR with a higher MAE in extreme wet and dry
Seasons 1 and 2 (MAE ˜ 43 kg N ha−1). Seasons 1 and 4 had the highest ac -curacy for Yield_N0 and Season 5 had the highest ac-curacy for EONR (MAE ˜ 29 kg N ha−1, Table 2). The reduced model performed the best for EONR and Yield_N0 for the dry Season 2 (MAE 49 kg N ha−1) and EONR predic
-tion error was the same as the full model in the wet Season 1(Table 2). Pre
-liminary testing of the EONR models using Season 6 resulted on an aver -age MAE of 52 kg N ha−1. The most accurate EONR predictions were found in landscape position with topsoil sand content<65% and SOM > 2.5% (ab
-solute difference between observed and predicted values was on average
17 kg N ha−1). The least accurate EONR predictions were for a sandy hill site with OM < 2%. Regarding YEONR prediction, the MAE ranged from 1 to 1.9
Mg ha−1 in most cases (Figure S7).
Fig. 5. Relative importance of the static and dynamic variables included in the re-duced models. Left panels shows variable included in the EONR model, middle in the YEONR model, and right in the Yield_N0 model.
4. Discussion
4.1. The economic optimum N rate in Argentina
Findings from this 51 N trial study suggest that there is potential for yield
increases in Argentina by improving N management. In 70% of our N
tri-als, the EONR was higher than the regional average N application rate cur
-rently used in Argentina (Bolsa de Cereales, 2018) and the average YEONR
was 4 Mg ha−1 higher than the regional average corn yield (Andrade and Satorre, 2015; Fig. 2). However, increasing N fertilization rate adds a signif -icant economic risk that most producers in this region were not willing to
accept because of the year-to-year variability in EONR (Puntel et al., 2017). Here we not only quantified the variability in EONR (Fig. 2)—a critical infor -mation that was previously missing for this region—but more importantly, we developed predictive models that account for weather and soil
variabil-ity to aid producer’s N decision-making. 4.2. Implications of the developed models
The developed models are field specific and fill the gap between very sim
-ple EONR prediction tools (one input) and more com-plex N tools such as simulations models that require several input parameters (Basso and Liu, 2018). In a recent review by Morris et al. (2018) on N recommendation tools,
it was noticeable that all current approaches used are either too simple or
too complex. Our models integrate key static and dynamic variables, reflect
-ing exist-ing knowledge on factors affect-ing EONR (e.g. precipitation distri
-bution), and has tremendous potential in view of increasing data availability in agriculture. While complex simulation models have already proven to be well-suited to support N decisions in specific fields (Basso et al., 2011) their utilization is still low because of challenges related to site-specific model calibration, large number of input requirements, and social difficulties in de
-livering model-based results to stakeholders (Salo et al., 2016; Banger et al., 2017; He et al., 2017). Our tool uses fewer inputs than a simulation model
and is easier to deploy across a broad geographic area. Compared to sim-ple N recommendation tools such as the yield goal approach, soil nitrate-N
test, or use of the long-term average EONR, our model accounts for spatial
and temporal variability and thus is well suited for application in precision agriculture.
This study expands earlier efforts to predict corn’s EONR using more than one explanatory variable in Argentina (Gregoret et al., 2011; Coyos et al., 2018) and in the USA (Qin et al., 2018). Gregoret et al. (2011) considered to
soil type, while Qin et al. (2018) considered soil water holding capacity and minimum groundwater table depth. Our model has six explanatory variables for EONR, which resulted from statistical analysis of their importance (Fig. 4). In agreement with Gregoret et al. (2006), precipitation was one of the
most important variables in our model. Importantly, our analysis revealed
other important factors such as soil depth or residue amount (Table 3) have been ignored in previous N tools. Thus, beyond predictability of EONR, this study also offers guidance on future research investment and data collec -tion from N trials.
Compared to complex simulation models (Liu et al., 2013; Puntel et al., 2016; Sela et al., 2017; Yang et al., 2014), the prediction error of the full EONR
model is similar (MAE ˜ 30–40 kg N ha−1). The reduced EONR model that was
designed for N forecasting at planting time, had 16% less predictive accu-racy than the full model but this was expected given the fewer inputs are
utilized. Preliminary testing showed that both full and reduced models gave
reasonable results (Figure S7), but further testing is needed across multi
-ple environments to increase confidence. Here we developed the concept
and provided a list of important parameters to be measured in future trials. As new data becomes available the proposed models could be further
im-proved and expanded. The developed models/framework have future po
-tential as they can provide farmers with flexibility to adjust their N rates for their specific fields given local weather conditions, soil properties, and man
-agement that differ from field to field.
In the future, some labor-intensive input parameters that are currently
part of our models could be further simplified (Table 2). For example, the amount of residue (kg/ha) can be replaced by categorical classes such as very low, low, normal, high and very high amounts. We tested this con
-cept (data not shown) and we found that the EONR prediction accuracy re
-mained at similar levels compared to models listed in Table 2. Similarly, the initial soil moisture at planting could be replaced by classes. This approach
may enhance deployment of the models across environments but more work is needed to develop and test it. Lastly, other variables such as plant
density (Carlone and Russell, 1987) or genotype (Gambin et al., 2016) that are known to influence EONR could be incorporated into the models. Grain and fertilizer prices could also be added towards developing a more flexi -ble N decision tool.
During model development we faced several challenges that should be
considered in future studies. We found that the relative importance of static
variables selected for inclusion in the models changed based on which
dy-namic variables were considered. This is evident by the fact that different
variables were found to be important and thus included in the full and re-duced models. For example, when we removed data on soil texture and
precipitation distribution, we found that ECa and terrain parameters became
more important in the models (data not shown). Lastly, some of the factors accounted for in our models could influence yield in a non-linear manner,
while in our models we considered the relationships to be linear.
4.3. Analysis of factors affecting the EONR and yields
Our study provides for the first time a comparative analysis of static and dy
-namic factors influencing EONR, YEONR and Yield_N0 that was missing from the literature. From a list of 54 factors, statistical analysis identified 14 to be the most important variables for EONR, YEONR and Yield_N0 predictions in this region (Figs. 4 and 5). It is interesting to note that yield of the previ
-ous crop, currently used to develop management zones and N recommen
-dations (Kersebaum et al., 2002), was not an important factor. In contrast, we found residue amount to explain significant portions of the YEONR and EONR variability (Figs. 4 and 5). Therefore future research should focus more
on determining residue amount and decomposition dynamics rather than on precisely estimating past crop yields to inform future N recommendations.
Our results suggest that variable N rate recommendations based on pre
-vious yields and static variables such as texture and SOM could fail when weather conditions are wet or extremely wet, which agrees with other stud
-ies such as Cerrato and Blackmer, 1991) and Vanotti and Bundy, 1994a,b. For example, the yield response to N was almost double in wet compared to dry seasons (Fig. 3), this is in agreement with previous studies (Kyveryga et al., 2009). The yield response to N was 3-fold higher in fine than coarse textured soils in dry and normal seasons, similar to Shahandeh et al. (2011). These examples strengthen the need for having a N recommendation tool
that combines temporal and spatial variability.
In this study, seasons 4 and 5 had high yields (˜12 Mg ha−1) with relatively low EONR (˜70 kg N ha−1; Fig. 3) and thus it is important to understand the cause. Analysis of different factors indicated that the reason was the distri -bution of precipitation within these two seasons: high frequency of small rain
events and low frequency of extreme rain events (> 20 mm) where N losses are likely to occur (Davis et al., 2000; Rimski-Korsakov et al., 2004). The high frequency of small rain events in Season 4 and 5 resulted in higher Yield_ N0, and reduced the yield response to N, therefore, decreasing the EONR (Fig. 3; Sogbedji et al., 2001).
Among different ways of analyzing precipitation data, we found that
the number of days with precipitation greater than 20mm from planting to silking, and from planting to harvest to be very good predictors of the
EONR and Yield_N0 variability in the full model (Fig. 4). Although there is
time-periods), our statistical analysis revealed no significant auto-correlation.
In the reduced model, precipitation events prior to planting, and soil
mois-ture relative to field capacity at planting, were the most important variables. Water table depth was a significant factor in the reduced Yield_N0 model
(Fig. 5) but not in the other models. Presumably this is because of its re
-lation to precipitation and landscape position (Gleeson et al., 2011). With the difficulty in measuring water tables, use of elevation position may be a good alternative because of the correlation between these variables (Fig. 4; Nosetto et al., 2009). Interestingly, relative elevation (Table 1) was one of the most important explanatory variables in both YEONR and Yield_N0 mod
-els. Other static factors such as SCA, pcurv, and slope are relevant in
regres-sion models, mainly because of their control on water availability (Van Itter
-sum et al., 2002), N dynamics, and thus yield (Kaspar et al., 2004; Kravchenko and Bullock, 2000).
Neither soil nitrate-N at planting nor the YEONR were good single pre
-dictors of the EONR (Fig. 3 to 6), this is in line with literature from rain
-fed regions (Lory and Scharf, 2003; Sawyer et al., 2006; Vanotti and Bundy, 1994a,b). In contrast, when soil nitrate information was combined with other
static and dynamic factors, it proved to be a very important factor in the
full Yield_N0 model and in the reduced EONR model (Figs. 4 and 5). This
further justifies our previous statement for the need to combined different approaches/data inputs towards improving predictability of EONR (Kay et al., 2006f).
Finally, we found a significant relationship between the difference of YEONR and Yield_N0 and the EONR (RAdj2 = 0.92; Fig. 3). This relationship is interesting because it offers an alternative way to estimate EONR using in
-formation on optimum and minimum yield. In a previous study (Puntel et al., 2016, 2018), we found that the APSIM crop model predicted optimum and minimum yields more accurately than EONR. This relationship may help in the way we currently calculate EONR via crop models. Furthermore, use
of this relationship will decrease the number of crop model simulations
re-quired to calculate the EONR (from 5 to 30 simulations to only two; Pun
-tel et al., 2016). 5. Conclusions
Our approach provides a new avenue to integrating and analyzing various
datasets towards development of data-driven recommendations to
grow-ers. These multifaceted datatypes are likely to become readily available in the near future through advances in technology. The new N models devel
N tools. The prediction accuracy was satisfactory but further testing across environments is needed to increase confidence. Beyond predictability, this study also offers guidance on which variables to be measured in future N
trials based on their importance.
Acknowledgments — This work was part of the Agriculture and Food Research Ini -tiative Hatch project No. 1004346 and partially funded by the Iowa State University
Plant Sciences Institute. We also thank the support of the farm’s owners where the trials were established, Doña Norma Pedemonte, German Molea, and Juan Carlos Pagani. Special thanks to Dr. Philip Dixon and the PhD student Kathleen Rey from
Iowa State University for their contribution in the statistical analysis and to Dr.
Mar-tinez-Feria for his helpful comments. Thanks also to Geronimo and Alcidez Gonza
-lez for hand harvesting the corn trials and the lab technicians at Clarion Soil Test
-ing Lab for analyz-ing the soil samples.
Appendix A. Supplementary data — Supplementary material follows the Refer
-ences; and the Excel model is attached to the main repository record for this article.
References
Abendroth, L.J., Elmore, R.W., Boyer, M.J., Marlay, S.K., 2011. “Corn Growth and
Development.” Publ. PMR 1009. Iowa State University Extension and Outreach,
Ames, IA.
Albarenque, S.M., Basso, B., Caviglia, O.P., Melchiori, R.J.M., 2016. Spatio-temporal nitrogen fertilizer response in maize: field study and modeling approach. Agron. J. 108 (5), 2110–2122.
Alexandratos, N., Bruinsma, J., 2012. World agriculture towards 2030/2050. Land Use Policy 20 (4), 375.
AMIS, 2018. Agricultural Market Information System. Retrieved from http://www. amisoutlook.org/amis-about/en/
Andrade, J.F., Satorre, E.H., 2015. Single and double crop systems in the argentine pampas: environmental determinants of annual grain yield. Field Crops Res.
177, 135–147.
Andrade, F.H., Uhart, S.A., Frugone, M., 1993. Intercepted radiation at flowering and kernel number in maize: shade versus plant density effects. Crop Sci. 33,
482–485.
Andraski, T.W., Bundy, L.G., Brye, K.R., 2000. Crop management and corn nitrogen rate effects on nitrate leaching. J. Environ. Qual. 29, 1095–1103.
Aparicio, V., Costa, J.L., Zamora, M., 2008. Nitrate leaching assessment in a
long-term experiment under supplementary irrigation in humid Argentina. Agric. Water Manag. 95 (12), 1361–1372.
Archontoulis, Sotirios V., Miguez, Fernando E., 2015. Nonlinear regression models and applications in agricultural research. Agron. J. 107 (2), 786–798.
Baethgen, W.E., Taylor, D.B., Alley, M.M., 1989. Quadratic programming method for
determining optimum nitrogen rate for winter wheat during tillering. Agron. J. 81, 557–559.
Banger, K., Yuan, M., Wang, J., Nafziger, E.D., Pittelkow, C.M., 2017. A vision for incorporating environmental effects into nitrogen management decision support tools for U.S. Maize production. Front. Plant Sci. 8, 1270.
Basso, B., Liu, L., 2018. Seasonal crop yield forecast: methods, applications, and accuracies. Advances in Agronomy. https://doi.org/10.1016/
bs.agron.2018.11.002
Basso, B., Ritchie, J.T., Pierce, F.J., Braga, R.P., 2001. Spatial validation of crop
models for precision agriculture. Agric. Syst. 68.
Basso, B., Ritchie, J.T., Cammarano, D., Sartori, L., 2011. A strategic and tactical management approach to select optimal N fertilizer rates for wheat in a spatially variable field. Eur. J. Agron. 35 (4), 215–222.
Basso, B., Cammarano, D., Fiorentino, C., Ritchie, J.T., 2013. Wheat yield response to spatially variable nitrogen fertilizer in Mediterranean environment. Eur. J.
Agron. 51, 65–70.
Brevik, E.C., Fenton, T.E., Lazari, A., 2006. Soil electrical conductivity as a function of soil water content and implications for soil mapping. Precis. Agric. 7 (6),
393–404.
Bolsa de Cereales, 2018. Relevamiento de Tecnologia Agricola Aplicada de la Bolsa de Cereales. Informe ReTAA No7. http://www.bolsadecereales.com/retaa Bullock, D.G., Bullock, D.S., 1994. Quadratic and quadratic-plus-plateau models
for predicting optimal nitrogen rate of corn: a comparison. Agron. J. 86 (1),
191–195.
Cabrera, M.L., Kissel, D.E., Vigil, M.F., 2005. Nitrogen mineralization from organic residues: research opportunities. J. Environ. Qual. 34, 75–79.
Carlone, M.R., Russell, W.A., 1987. Response to plant densities and nitrogen levels for four maize cultivars from different eras of breeding 1. Crop Sci. 27 (3),
465–470.
Cerrato, M.E., Blackmer, A.M., 1990. Comparison of models for describing; corn
yield response to nitrogen fertilizer. Agron. J. 82 (1), 138–143.
Cerrato, M.E., Blackmer, A.M., 1991. Relationships between leaf nitrogen concentrations and the nitrogen status of corn. J. Prod. Agric. 4 (4), 525–531. Chang, J., Clay, D., Carlson, D.E., Reese, C.G., Clay, C.L., Ellsbury, S.A., M. M, 2004. Defining yield goals and management zones to minimize yield and nitrogen and phosphorus fertilizer recommendation errors. Agron. J. 3, 825–831. Coyos, T., Borrás, L., Gambin, B.L., 2018. Site-specific covariates affecting yield
response to nitrogen of late-sown maize in Central Argentina. Agron. J. 110 (4),
1544–1553.
Davis, C.W., 1916. Studies on the phenoldisulphonic acid method for determining
nitrates in soils. J. Indus. Eng. Chem. 9, 290–295.
Davis, D.M., Gowda, P.H., Mulla, D.J., Randall, G.W., 2000. Modeling Nitrate
Nitrogen Leaching in Response to Nitrogen Fertilizer Rate and Tile Drain Depth or Spacing for Southern Minnesota, USA. J. Environ. Qual. 29, 1568–1581. Dhital, S., Raun, W.R., 2016. Variability in optimum nitrogen rates for maize. Agron.
Dinnes, D.L., Karlen, D.L., Jaynes, D.B., Kaspar, T.C., Hatfield, J.L., 2002. Review and
interpretation: Nitrogen management strategies to reduce nitrate leaching in
tile-drained midwestern soils. Agron. J. 94 (1), 153–171.
Edreira, J.R., Otegui, M.E., 2013. Heat stress in temperate and tropical maize hybrids: a novel approach for assessing sources of kernel loss in field conditions. Field Crops Res. 142, 58–67.
ESRI, 2018. ArcGIS Desktop: Release 10.6.1 Redlands. Environmental Systems Research Institute., CA.
Ferrari, M., Castellarin, J., Sainz Rozas, H., Vivas, H., Melchiori, R., y Gudelj, V., 2010. Evaluación de métodos de diagnóstico de fertilidad nitrogenada para el cultivo de trigo en la región pampeana. Inform. Agron. 46, 10–12 IPNI.
Gambin, B.L., Coyos, T., Di Mauro, G., Borrás, L., Garibaldi, L.A., 2016. Exploring genotype, management, and environmental variables influencing grain yield of late-sown maize in central Argentina. Agric. Syst. 146, 11–19.
Gleeson, T., Marklund, L., Smith, L., Manning, A.H., 2011. Classifying the water table at regional to continental scales. Geophys. Res. Lett. 38 (5), 1–6. Gregoret, M.C., Dardanelli, J., Bongiovanni, R., Díaz-Zorita, M., 2006. Modelo de
respuesta sitio-específica del maíz al nitrógeno y agua edáfica en un haplustol. Ciencia del Suelo 24 (2), 147–159.
Gregoret, M.C., Díaz-Zorita, M., Dardanelli, J., Bongiovanni, R., 2011. Regional model for nitrogen fertilization of site-specific rainfed corn in haplustolls of the
central Pampas, Argentina. Precis. Agric. 12, 831–849. https://doi.org/10.1007/ s11119-011- 9224-7
Gromping, Ulrike, 2006. Relative Importance for Linear Regression in R : The Package Relaimpo 17 1.
He, D., Wang, E., Wang, J., Robertson, M.J., 2017. Data requirement for effective calibration of process-based crop models. Agric. For. Meteorol. 234, 136–148. Heiniger, R.W., McBride, R.G., Clay, D.E., 2003. Using soil electrical conductivity to
improve nutrient management. Agron. J. 95 (3), 508–519.
Hergert, G.W., Ferguson, R.B., Shapiro, C.A., Penas, E.J., Anderson, F.B., 1995.
Classical statistical and geostatistical analysis of soil nitrate-N spatial variability.
In Site-specific Management for Agricultural Systems, pp. 175–186 American
Society of Agronomy, Crop Science Society of America, Soil Science Society of America.
Jaynes, Dan B., 2011. Confidence bands for measured economically optimal nitrogen rates. Precis. Agric. 12 (2), 196–213.
Jaynes, D.B., 2012. Changes in Yield and Nitrate Losses from Using Drainage
Water Management in Central Iowa, United States. J. Soil Water Conserv. 67 (6),
485–494.
Kaspar, T.C., Kaspar, T.C., Pulido, D.J., Fenton, T.E., Colvin, T.S., Karlen, D.L., Jaynes, D.B., Meek, D.W., 2004. Relationship of corn and soybean yield to soil and terrain properties. Agron. J. 96 (3), 700–709.
Kay, B.D., Mahboubi, A.A., Beauchamp, E.G., Dharmakeerthi, R.S., 2006. Integrating Soil and Weather Data to Describe Variability in Plant Available Nitrogen. Soil Sci. Soc. Am. J. 70 (4), 1210–1221.
Kersebaum, K.C., Lorenz, K., Reuter, H.I., Wendroth, O., Ahuja, L.R., 2002. Modelling
variability. Agricultural System Models in Field Research and Technology Transfer. CRC Press, pp. 234–257.
King, J.A., Dampney, P.M.R., Lark, R.M., Wheeler, H.C., Bradley, R.I., Mayr, T.R., 2005. Mapping potential crop management zones within fields: use of yield-map series and patterns of soil physical properties identified by electromagnetic induction sensing. Precis. Agric. 6 (2), 167–181.
Kitchen, N.R., Drummond, S.T., Lund, E.D., Sudduth, K.A., Buchleiter, G.W.,
2003. Soil electrical conductivity and topography related to yield for three contrasting soil-crop systems. Agron. J. 95, 483–495.
Kitchen, N.R., Sudduth, K.A., Myers, D.B., Drummond, S.T., Hong, S.Y., 2005. Delineating productivity zones on claypan soil fields using apparent soil electrical conductivity. Comput. Electron. Agric. 46 (1-3), 285–308.
Kravchenko, A.N., Bullock, D.G., 2000. Correlation of corn and soybean grain yield with topography and soil properties. Agron. J. 92 (1), 75–83.
Kyveryga, P.M., Blackmer, A.M., Zhang, J., 2009. Characterizing and classifying variability in corn yield response to nitrogen fertilization on subfield and field
scales. Agron. J. 101, 269–277. https://doi.org/10.2134/agronj2008.0168.
LECO, 2008. Organic application notes. LECO, St. Joseph, MI. Available at http://
www.leco.com/resources/application_notes/OrganicCHNSOappNoteForm.html
[accessed 18 Feb. 2008; verified 16 Dec. 2011].
Liu, S., Yang, J.Y., Zhang, X.Y., Drury, C.F., Reynolds, W.D., Hoogenboom, G., 2013.
Modelling crop yield, soil water content and soil temperature for a
soybean-maize rotation under conventional and conservation tillage systems in Northeast China. Agric. Water Manag. 123, 32–44.
Lory, J.A., Scharf, P.C., 2003. Yield goal versus Delta yield for predicting fertilizer nitrogen need in corn. Agron. J. 95 (4), 994–999.
Mamo, M., Malzer, G.L., Mulla, D.J., Huggins, D.R., Strock, J., 2003. Spatial and
temporal variation in economically optimum nitrogen rate for corn. Agron. J.
95 (4), 958–964.
Mejia, M.N., Madramootoo, C.A., Broughton, R.S., 2000. Influence of water table management on corn and soybean yields. Agr. Water Manage 46, 73–89. Melchiori, R.J., Novelli, L.E., Gregorutti, V.C., Caviglia, O.P., 2014. Stover quality and
soil organic carbon in long-term nitrogen-fertilized maize. Agron. J. 106 (5),
1709–1716.
Miller, B.A., 2014. Semantic calibration of digital terrain analysis. Cartogr. Geogr.
Inf. Sci. J. 41, 166–176. https://doi.org/10.1080/15230406.2014.883488.
Moore, I.D., Grayson, R.B., 1991. Terrain based prediction of runoff with vector elevation data. Wat. Resour. Res. 27, 1177–1191.
Morris, T.F., Murrell, T.S., Beegle, D.B., Camberato, J.J., Ferguson, R.B., Grove, J., Ketterings, Q., Kyveryga, P.M., Laboski, C.A.M., McGrath, J.M., Meisinger, J.J., Melkonian, J., Moebius-Clune, B., Nafziger, E.D., Osmond, D., Sawyer, J.E., Scharf, P.C., Smith, W., Spargo, J.T., van Es, H.M., Yang, H., 2018. Strengths and
limitations of nitrogen rate recommendations for corn and opportunities for improvement. Agron. J. 110, 1–137.
Nosetto, M.D., Jobbágy, E.G., Jackson, R.B., Sznaider, G.A., 2009. Reciprocal influence of crops and shallow ground water in sandy landscapes of the inland pampas. Field Crops Res. 113 (2), 138–148.
Ordóñez, R.A., Savin, R., Cossani, C.M., Slafer, G.A., 2015. Yield response to heat stress as affected by nitrogen availability in maize. Field Crops Res. 183, 184–203. Pagani, A., Echeverria, H.E., Sainz Rozas, H., Barbieri, P.A., 2008. Dosis óptima
económica de nitrógeno en maíz bajo siembra directa en el Sudeste bonaerense. Ciencia del Suelo 26 (2), 179–188.
Peralta, N.R., Costa, J.L., Balzarini, M., Angelini, H., 2013. Delineation of
management zones with measurements of soil apparent electrical conductivity in the southeastern pampas. Can. J. Soil Sci. 93 (2), 205–218.
Puntel, L.A., Sawyer, J.E., Barker, D.W., Dietzel, R., Poffenbarger, H., Castellano, M.J., Moore, K.J., Thorburn, P., Archontoulis, S.V., 2016. Modeling long-term corn yield response to nitrogen rate and crop rotation. Front. Plant Sci. 7 November
2016.
Puntel, L., Archontoulis, S.A., Pagani, A., 2017. Site-Specific Nitrogen Response Functions for Maize in Argentina.” in American Society of Agronomy. Tampa, FL. Retrieved.
https://scisoc.confex.com/crops/2017am/webprogram/Paper107787.html
Puntel, L.A., Sawyer, J.E., Barker, D.W., Thorburn, P.J., Castellano, M.J., Moore, K.J.,
VanLoocke, A., Heaton, E.A., Archontoulis, S.V., 2018. A systems modeling
approach to forecast corn economic optimum nitrogen rate. Front. Plant Sci. 9. Qin, Z., Myers, D.B., Ransom, C.J., Kitchen, N.R., Liang, Sang-Zi, Camberato, J.J.,
Carter, P.R., Ferguson, R.B., Fernandez, F.G., Franzen, D.W., Laboski, C.A.M., Malone, B.D., Nafziger, E.D., Sawyer, J.E., Shanahan, J.F., 2018. Application of Machine Learning Methodologies for Predicting Corn Economic Optimal Nitrogen Rate. Agron. J in press.
R Core Team, 2018. R: A Language and Environment for Statistical Computing. Retrieved (http://www.r-project.org/).. .
Rimski-Korsakov, H., Rubio, G., Lavado, R.S., 2004. Potential Nitrate Losses under Different Agricultural Practices in the Pampas Region, Argentina. Agric. Water Manag. 65 (2), 83–94.
Ruiz, R., Satorre, E., Maddoni, G., Carcova, J., Otegui, M., 2001. Umbrales de decisión para la fertilización nitrogenada en maíz. VII Congreso Nacional de Maíz. Sainz Rozas, H., Echeverría, H., Studdert, G., Dominguez, G., 2000. Evaluation of
the presidedress soil nitrogen test for no-tillage maize fertilized at planting. Agron. J. 92 (6), 1176–1183.
Sainz Rozas, H., Echeverría, H., Herfurt, E., Studdert, E., 2001. Nitrato en la base del tallo de maíz. II. Diagnóstico de la nutrición nitrogenada. Ciencia del Suelo 19 (2), 125–135.
Salo, T.J., Palosuo, T., Kersebaum, K.C., Nendel, C., Angulo, C., Ewert, F., Bindi, M., Calanca, P., Klein, T., Moriondo, M., Ferrise, R., 2016. Comparing the
performance of 11 crop simulation models in predicting yield response to
nitrogen fertilization. J. Agric. Sci. 154 (7), 1218–1240.
Satorre, E., Mercau, J., 2001. Bases de decisión para la fertilización nitrogenada en maíz. Informe Convenio AACREA-FAUBA.
Sawyer, J., Nafziger, E., Randall, G., Bundy, L., Rehm, G., Joern, B., 2006. Concepts and Rationale for Regional Nitrogen Rate Guidelines for Corn Concepts and Rationale for Regional Nitrogen Rate Guidelines for Corn. Iowa State University, University Extension (April 2006). pp. 1–28.
Saxton, K.E., Rawls, W.J., 2006. Soil water characteristic estimates by texture and organic matter for hydrologic solutions. Soil Sci. Soc. Am. J. 70 (5), 1569–1578. Scharf, Peter C., 2015. “Managing Nitrogen” Managing Nitrogen in Crop
Production. American Society of Agronomy Inc., USA, pp. 25–76.
Scharf, P.C., Lory, J.A., 2006. Best Management Practices for Nitrogen Fertilizer in Missouri. University of Missouri Extension. http://plantsci.missouri.edu/ nutrientmanagement/nitrogen/practices.htm [last accessed 19 March 2013]. Sela, S., van Es, H.M., Moebius-Clune, B.N., Marjerison, R., Moebius-Clune,
D., Schindelbeck, R., 2017. Dynamic model improves agronomic and environmental outcomes for maize nitrogen management over static approach. J. Environ. Qual. 46 (2), 311.
Shahandeh, H., Wright, A.L., Hons, F.M., 2011. Use of soil nitrogen parameters and texture for spatially-variable nitrogen fertilization. Precis. Agric. 12 (1), 146–163. Shaner, D.L., Khosla, R., Brodahl, M.K., Buchleiter, G.W., Farahani, H.J., 2008. How
Well Does Zone Sampling Based on Soil Electrical Conductivity Maps Represent Soil Variability? Agron. J. 100 (5), 1472–1480.
Shapiro, C.A., Ferguson, R.B., Hergert, G.W., Wortmann, C.S., Walters, D.T., 2008. Fertilizer Suggestions for Corn. EC117, Nebraska Extension, Lincoln, NE, pp. 1–6. Sogbedji, J., Sogbedji, J.M., van ES, H.M., Klausner, S.D., Bouldin, D.R., Cox, W.J.,
2001. Spatial and temporal processes affecting nitrogen availability at the landscape scale. Soil Tillage Res. 58 (3–4), 233–244.
Soil Survey Staff, 2014. Illustrated Guide to Soil Taxonomy. U.S. Department of Agriculture, Natural Resources Conservation Service. National Soil Survey
Center, Lincoln, Nebraska.
Stanford, G., Legg, J., Smith, S., 1973. Soil nitrogen availability evaluations based on nitrogen mineralization potentials of soils and uptake of labeled and unlabeled nitrogen by plants. Int. J. on Plant-Soil Relationships 39 (1), 113–124. Tremblay, N., Bouroubi, M.Y., Vigneault, P., Bélec, C., 2011. Guidelines for
in-season nitrogen application for maize (Zea Mays l.) based on soil and terrain properties. Field Crops Res. 122, 157–182.
Van Ittersum, M.K., Leffelaar, P.A., Van Keulen, H., Kropff, M.J., Bastiaans, L.,
Goudriaan, J., 2002. Developments in modelling crop growth, cropping systems and production systems in the Wageningen school. NJAS—Wageningen J. Life Sci. 50 (2), 239–247.
Vanotti, M.B., Bundy, L.G., 1994a. An alternative rationale for corn nitrogen fertilizer recommendations. J. Prod. Agric. 7 (2), 243–249.
Vanotti, M.B., Bundy, L.G., 1994b. Corn nitrogen recommendations based on yield response data. J. Prod. Agric. 7 (2), 249.
Wang, D., Anderson, D.W., 1998. Direct measurement of organic carbon content in soils by the Leco CR-12 carbon analyzer. Commun. Soil Sci. Plant Anal. 29 (1-2),
15–21.
Wilson, J.P., Gallant, J.C., 2000. Digital terrain analysis. Terrain analysis: principles and applications 6 (12), 1–27.
Yang, J.M., Yang, J.Y., Liu, S., Hoogenboom, G., 2014. An evaluation of the statistical
methods for testing the performance of crop models with observed data. Agric. Syst. 127, 81–89.
Figure S1.
Elevation, apparent electrical conductivity at 90 cm depth (ECa), plan curvature, and slope
maps. Purple flags show an example of the location of some trials.
for selected periods (bottom panel). Symbols used in the top panel, P, S, and H indicate planting, silking, and
harvest time, respectively. In the bottom panel, numbers above the columns indicate number of precipitation
events
area (SCA, pixel); percent of slope; relative elevation (Rel_elev, meters); plan curvature (degrees). Density plots of
the variables are shown in the diagonal.
matter at 20 cm (Sand_20 and OM_20), soil nitrate (N_0_60, kg N ha
), and apparent electrical conductivity at 90 cm
depth (ECa_90). Density plots of the variables are shown in the diagonal.
(EONR). The value (ratio) of 10 was used in this study.
Figure S7.
Predicted versus observed economic optimum N rate (EONR), yield at EONR (YEONR), and
yield at non fertilizer (Yield_N0) for the validation data set using full model (circles) and reduced model
(triangles). Diagonal dashed line shows 1:1 relationship. MAE is the mean absolute error.
Table S1. Coefficient of determination (R2) for the fitting of 51 yield response to nitrogen trials.
Trial ID R2 Model 1 0.61 Quadratic-plateu 2 0.67 Quadratic-plateu 3 0.87 Quadratic-plateu 4 0.86 Quadratic-plateu 5 0.91 Quadratic-plateu 6 0.87 Quadratic-plateu 7 0.90 Quadratic-plateu 8 0.85 Quadratic-plateu 9 0.91 Quadratic-plateu 10 0.78 Quadratic-plateu 11 0.60 Quadratic-plateu 12 0.87 Quadratic-plateu 13 0.61 Quadratic-plateu 14 NA NA 15 NA NA 16 0.86 Quadratic-plateu 17 0.55 Quadratic-plateu 18 0.70 Quadratic-plateu 19 NA NA 20 0.78 Quadratic-plateu 21 NA NA 22 0.61 Quadratic-plateu 23 0.44 Quadratic-plateu 24 0.50 Quadratic-plateu 25 0.69 Quadratic 26 0.76 Quadratic-plateu 27 0.83 Quadratic-plateu 28 0.73 Quadratic-plateu 29 NA NA 30 NA NA 31 0.48 Quadratic-plateu 32 0.50 Quadratic-plateu 33 0.94 Quadratic-plateu 34 0.52 Quadratic-plateu 35 0.83 Quadratic-plateu 36 0.65 Quadratic 37 NA NA 38 0.85 Quadratic-plateu 39 0.75 Quadratic-plateu 40 0.69 Quadratic-plateu 41 0.55 Quadratic-plateu 42 0.85 Quadratic-plateu 43 0.89 Quadratic-plateu 44 0.65 Quadratic-plateu47 NA NA
48 NA NA
49 0.87 Quadratic-plateu
50 0.78 Quadratic-plateu
51 NA NA
